A vision-language model read each image together with its QA and judged the item. The pass is not a filter that only removes rows — it rewrites answers it finds wrong but salvageable, drops what it cannot salvage, and adds QA where the image supports more than the source provided. Each row carries the judge's own record in clean_meta, including the cleaner identity, the policy it applied, and its per-item scores for legibility, richness and coverage.

A large share of the QA here was written by the cleaning pass, not by the original dataset. Across the pool that share runs from roughly half to over 80% of a family's turns, and it is reported in the table above. Those rows have an empty org_answer because no pre-clean original exists. Treat them as model-generated supervision: they were judged against the image, but they are not human annotation, and model-written QA is where formatting defects are most likely.

The effect on the answers that were carried over is substantive rather than cosmetic. In this pool the pass corrects values, not just wording — a curve's critical point restated from 4.00 to 2.00, a computed ratio from 1 to Approximately 1.33 — and for some families it removed the content entirely, which is why those families are absent here rather than published empty.

`answer` vs `org_answer`

answer — the cleaned answer. Train on this.
org_answer — the pre-cleaning answer from the same (image, question) in the uncleaned pool. It is empty for QA the cleaning pass added, which have no original.

Where the two differ, the difference is the correction. Keeping both makes every edit auditable instead of asking you to take the cleaning on trust.

This dataset supersedes an earlier upload of the same family that was built from the pre-clean pool. That version carried the uncorrected answers and has been deleted.

Schema

image — HF Image(), renders directly in the dataset viewer
image_sha256 — content hash of the image
n_qa — number of turns attached to this image
qa — list of {question, answer, org_answer}, in source order
family / clean_meta — provenance and the judge's per-row record

A row with more than one entry in qa is a conversation over a single image, not a set of independent questions — the turns depend on each other and the image belongs to all of them. Keep them together and in order when training.

Parquet row groups are written at ~32 MB so the dataset viewer pages without stalling.

What is NOT claimed

No rejection sampling was run against this family: there is no accept/reject verdict per row, and answer is the cleaned reference rather than a model generation that a string verifier confirmed. For that, use the *_RS_think / *_rejected datasets in this org.

Examples

10 rows taken straight from this dataset. Images are the original files as stored here - no downscaling, no recompression - and every turn is shown in full.

Example 1 — 7 turns on one image

#	question	answer
1	Is there a table in this image?	Yes, I can see a table in the image.
2	Is there a staircase in this image?	No, there is no staircase visible in this image.
3	What large electronic device is mounted on the wall on the left side?	A flat-screen television.
4	What color are the walls in the main living area?	Yellow.
5	Is there a person visible in the image?	Yes, a woman is standing in the kitchen area in the background.
6	What feature is built into the wall on the far right?	A fireplace.
7	What type of flooring is visible in the room?	Wooden floorboards.

Example 2 — 5 turns on one image

#	question	answer
1	Is there a cat in this image?	Yes, I can see a cat in the image.
2	Is there a dog in this image?	No, there is no dog visible in this image.
3	What object is the cat lying behind?	The cat is lying behind a silver laptop computer.
4	What are the main colors of the cat's fur?	The cat has black and white fur.
5	What color are the cat's eyes?	The cat has yellow or greenish-yellow eyes.

Example 3 — 5 turns on one image

#	question	answer
1	Is there snow in this image?	Yes, I can see a snow in the image.
2	Is there a body of water in this image?	No, there is no body of water visible in this image.
3	What large natural feature dominates the background of the image?	A rugged mountain peak covered in snow with exposed rock faces.
4	What group of people is visible in the lower center of the image?	A group of skiers standing together.
5	What curved lines are visible in the snow on the right side?	Ski tracks leading up the slope.

Example 4 — 5 turns on one image

#	question	answer
1	What might be the reason for the unusual vantage point of this image?	To emphasize the perspective of someone using the toilet..
2	What is covering the toilet seat?	A fuzzy, textured cover.
3	What is visible inside the toilet bowl?	Water with bubbles, indicating a flush.
4	What body parts are visible at the bottom of the frame?	A person's feet wearing dark shoes and pants.
5	What object is hanging on the wall to the right of the toilet?	A dark garment or bag with white markings.

Example 5 — 6 turns on one image

#	question	answer
1	Does this image describe a historical building?	Looking at this thermal image, Yes, this image shows a historical building.
2	Does this image describe a rural landscape?	No, this image does not show a rural landscape.
3	How many people are visible in the foreground?	Two people are visible standing in the foreground.
4	What is the color scheme of the image?	The image is in black and white (monochrome).
5	What architectural features are prominent on the building?	The building features multiple towers, domes, and arched windows.
6	What objects are visible in the plaza area in front of the building?	Several ornate street lamps and a fountain structure are visible in the plaza.

Example 6 — 4 turns on one image

#	question	answer
1	What could be the likely reason for capturing this image?	Looking at this thermal image, To monitor human activity in a specific area..
2	What imaging modality is depicted in this picture?	Thermal imaging.
3	How many distinct human figures are walking in the foreground?	Two.
4	What large features are visible in the background behind the people?	Large rock formations or hills.

Example 7 — 5 turns on one image

#	question	answer
1	Does this image describe a person walking in an outdoor setting?	Looking at this thermal image, Yes, this image shows a person walking in an outdoor setting.
2	Does this image describe a busy urban environment?	No, this image does not show a busy urban environment.
3	What type of imaging technology is used to capture this scene?	Thermal imaging.
4	What large natural features are visible in the background?	Trees or bushes.
5	What is the color scheme of the image?	Grayscale (black and white).

Example 8 — 6 turns on one image

#	question	answer
1	Does this image describe a place with water?	Looking at this thermal image, Yes, this image shows a place with water.
2	Does this image describe a densely populated urban area?	No, this image does not show a densely populated urban area.
3	What is the color scheme of the image?	The image is in black and white (grayscale).
4	What occupies the upper half of the image?	A dark, mostly featureless sky.
5	What features are visible in the immediate foreground?	A rough, textured ground surface, likely a beach or shoreline, with faint linear tracks and scattered light patches.
6	What is visible in the distance across the water?	A hazy landmass or shoreline.

Example 9 — 6 turns on one image

#	question	answer
1	Does this image depict a street sign indicating a neighborhood?	Yes, this image shows a street sign indicating a neighborhood.
2	Is there a clear view of public transportation in this image?	No.
3	What is the name written on the street sign?	Othello
4	What is the name of the restaurant visible in the background?	Cafe HUONG QUE
5	What phone number is listed on the sign above the restaurant entrance?	206-722-4318
6	Describe the logo on the left side of the street sign.	A white silhouette of a deer inside a blue circle.

Example 10 — 6 turns on one image

#	question	answer
1	Are there more than two people in this image?	Yes, there are more than two people in the image.
2	Are there less than three trees in this image?	No, there are not less than three trees in the image.
3	What white handheld gaming device is the person holding in front of their face?	A Nintendo DS Lite.
4	What text is visible on the strap hanging from the gaming device?	The words 'SONY' and 'HOLDS' are visible on the strap.
5	What color is the shirt worn by the person in the foreground?	Green.
6	What object is being held by one of the people walking in the background?	A white umbrella.

Downloads last month: 17

Total file size:

1.06 GB

You need to agree to share your Hugging Face information to access this dataset

spark_cleaned

How this was cleaned

answer vs org_answer